Comparison Of A Hybrid Neural Network And Semi-Distributed Simulator For Stream Flow Prediction

Lariyah Mohd Sidek, Milad Jajarmizadeh, Sobri Harun, Shamsuddin Shahid, Hidayah Basri

Research output: Contribution to conferencePaper

4 Downloads (Pure)


Hydrological models are widely used for the simulation of stream flow in order to aid
water resources planning and management in catchment or river basin. Numerous
hydrological models have been developed based on different theories. Performance of
such models depends on hydroclimatic
setting of a catchment. In the present study,
performance of a widely used physically based distributed model known as Soil and
Water Assessment (SWAT) and a datadriven
model, namely hybrid artificial neural
network (HANN), has been evaluated to simulate stream flow in an arid catchment
located in the south of Iran. Data related to topography, hydrometeorology, land cover,
and soil were collected and processed for this purpose. The models were calibrated and
validated with same time period to evaluate the advantage and disadvantages of
different models. The results showed SWAT outperformed HANN in terms of relative
errors such as NashSutcliffe
efficiency and percent of bias during model validation.
Other error indicates, namely root mean square error (RMSE), mean square error, and
mean relative error (MRE), were found close to zero for SWAT during both model
calibration and validation. The study suggests that both models have their own
promising flow prediction due to their own features and capabilities for daily flow.
Original languageEnglish
Number of pages12
Publication statusAccepted/In press - 10 Apr 2016


All Science Journal Classification (ASJC) codes

  • Modelling and Simulation

Cite this